When migrate_pages() steps on transparent huge page it splits the page and migrates only a head page. Therefore after migration NR_ISOLATED_ANON counter is dropped by one. When memcg_numa_isolate_pages() isolates huge page it increase NR_ISOLATED_ANON counter by the number of sub-pages in hugepage, which is never dropped by migrate_pages() as above. Eventually this leads to complete lockup, because every allocation will stuck in too_many_isolated() loop.
This patch fixes this by splitting huge page before migration and adding sub-pages to migration list. https://jira.sw.ru/browse/PSBM-54455 Fixes: a2d6b96306e3 ("mm: memcontrol: add memory.numa_migrate file") Signed-off-by: Andrey Ryabinin <[email protected]> --- mm/memcontrol.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/mm/memcontrol.c b/mm/memcontrol.c index cdf1b48..6366cc0 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -5740,12 +5740,13 @@ static long memcg_numa_isolate_pages(struct lruvec *lruvec, enum lru_list lru, { struct list_head *src = &lruvec->lists[lru]; struct zone *zone = lruvec_zone(lruvec); + struct page *page; long scanned = 0, taken = 0; spin_lock_irq(&zone->lru_lock); while (!list_empty(src) && scanned < nr_to_scan && taken < nr_to_scan) { - struct page *page = list_last_entry(src, struct page, lru); int nr_pages; + page = list_last_entry(src, struct page, lru); scanned++; @@ -5769,6 +5770,15 @@ static long memcg_numa_isolate_pages(struct lruvec *lruvec, enum lru_list lru, __mod_zone_page_state(zone, NR_ISOLATED_ANON + is_file_lru(lru), taken); spin_unlock_irq(&zone->lru_lock); + list_for_each_entry(page, dst, lru) { + if (PageTransHuge(page) && split_huge_page_to_list(page, dst)) { + list_del(&page->lru); + mod_zone_page_state(zone, NR_ISOLATED_ANON, + HPAGE_PMD_NR); + putback_lru_page(page); + } + } + return scanned; } -- 2.7.3 _______________________________________________ Devel mailing list [email protected] https://lists.openvz.org/mailman/listinfo/devel
